Multi-Classes of Microorganisms Classification Using Hybrid Deep Stacked Autoencoder

Abstract

Classification of microorganisms is critical for clinical and medical research; however, it remains challenging due to two key limitations: the limited availability of datasets of microscope images of microorganisms and the high morphological similarity between different classes. These limitations often result in misclassification and poor accuracy. To address these issues, the study proposes a hybrid model that combines a convolutional network and recurrent neural networks to extract more efficient features. This study evaluated the model on benchmark datasets. The model achieved 86% accuracy on the large-scale dataset and 97% accuracy on the small-scale dataset. This raises the probability of overfitting. For this, we utilized regularization and data augmentation techniques to reduce this risk. The main contribution is the development of a hybrid architecture that combines CNN, Bi-LSTM, and LSTM  to capture bidirectional temporal dependencies and spatial sequential patterns, and compares the results between the proposed model and baseline model, which used CNN, BILSTM, LSTM, CNN-BILSTM, and CNN-LSTM. This integration of the proposed model, which enhances feature extraction and provides a more robust learning compared to conventional CNN–LSTM or CNN–BiLSTM approaches. These results emphasize the model's potential to enhance diagnostic processes, reduce time and labor, and support global health efforts by alleviating the workload on biologists and specialists.